Sensitivity Analysis of Death Distribution Methods

1
Sensitivity Analysis of Death Distribution Methods

• Kenneth Hill
• Yoonjoung Choi

2
Key Assumptions

• Reporting of deaths invariant with age
• Reporting of population invariant with age
• Reporting of population invariant between
  censuses (SEG only)
• No age misreporting for survivors
• No age misreporting for deaths
• Closed population

3
Sensitivity Analysis

• Meets SEG assumptions
• Meets GGB assumptions
• Population coverage varies by age
• Death coverage varies by age
• Net migration

4
a) No data errors, observed 45q15 0.309
Growth Balance
Synthetic Extinct Generations
Intercept 0.0001 Slope 0.996 Adjusted
45q15 0.308
Average 15 to 55 1.000 Adjusted 45q15
0.309
5
b) Deaths omitted by 20, observed 45q15 0.256
Growth Balance
Synthetic Extinct Generations
Intercept 0.0001 Slope 1.245 Adjusted
45q15 0.308
Average 15 to 55 0.800 Adjusted 45q15
0.309
6
c) No age errors, 2 decline in coverage from
first to second census observed
45q15 0.312
Growth Balance
Synthetic Extinct Generations
Intercept 0.0042 Slope 0.986 Adjusted
45q15 0.308
Average 15 to 55 0.881 Adjusted 45q15
0.345
7
d) Age Misreporting, Population Deaths, no
Omission observed 45q15 0.313
Growth Balance
Synthetic Extinct Generations
Intercept 0.0003 Slope 0.912 Adjusted
45q15 0.290
Average 15 to 55 1.044 Adjusted 45q15
0.302
8
e) No Errors, Emigration 5 per 1,000
observed 45q15 0.309
Growth Balance
Synthetic Extinct Generations
Intercept 0.0064 Slope 0.942 Adjusted
45q15 0.294
Average 15 to 55 0.833 Adjusted 45q15
0.355
9
f) No age errors, 2 decline in coverage from
first to second census, emigration observed
45q15 0.311
Growth Balance
Synthetic Extinct Generations
Intercept 0.0105 Slope 0.933 Adjusted
45q15 0.294
Average 15 to 55 0.747 Adjusted 45q15
0.387
10
g) Age Errors, Omission of Deaths, Coverage
Decline observed 45q15 0.261
Growth Balance
Synthetic Extinct Generations
Intercept 0.0043 Slope 1.129 Adjusted
45q15 0.290
Average 15 to 55 0.733 Adjusted 45q15
0.339
11
Summary of Estimates
15 to 55
15 to 75
12
Conclusions

• If assumptions are met, both GGB and SEG work
  very well
• SEG sensitive to census coverage change
• Age misreporting does not have huge effects
• Varying coverage of population by age has big
  effect on GGB, less on SEG
• Net migration has larger effect on SEG than GGB
• 15 to 75 age range for fitting slightly better
• Combined method slightly preferable